基于数学形态学和局部均值分解的旋转机械故障诊断方法

发布时间：2018-05-06 02:26

本文选题：故障诊断 + 特征提取　；参考：《燕山大学》2015年博士论文

【摘要】：液压泵、滚动轴承和齿轮等旋转机械零部件在工业领域起着重要的传动作用,它们广泛地应用于冶金设备、工程机械、精密机床、精密仪器、航空航天、汽车、船舶和石油化工等重要工业领域。这些旋转机械零部件通常大都处于高温、高压和高速等恶劣的工作环境,这也就导致了它们的健康状态劣化速度比较快,具有较高的故障率。本文针对旋转机械零部件的故障诊断问题,研究了基于数学形态学和局部均值分解的方法。数学形态学和局部均值分解都是基于信号形态特征进行的分析,所以二者都具有高度自适应性和数据驱动性。本文主要进行了以下几个方面的研究工作:(1)针对扁平型结构元素长度选择的经验性和任意性问题,提出一种基于形态差值算子和功率谱熵相结合来确定扁平型结构元素的最优滤波长度的方法。采用不同扁平型结构元素长度的形态差值算子对信号进行滤波处理,对滤波后的信号计算功率谱熵,信号功率谱熵的最小值所对应的长度即为扁平型结构元素的最优滤波长度,并采用仿真信号和液压泵实测故障信号对所提方法的有效性进行验证。(2)利用一种基于形态差值算子和形态指数相结合的方法对以滑靴磨损故障为研究对象的液压泵健康状态评估的问题进行研究。采用形态差值算子对滑靴不同磨损程度的故障信号进行滤波处理,提取各种磨损程度下故障信号的形态指数,分析形态指数对滑靴磨损故障及其劣化程度的变化规律和敏感性,实现对滑靴磨损故障的诊断及其劣化程度的评估,并利用液压泵实测故障信号对所提方法的有效性进行验证。(3)利用一种基于形态差值算子和差分熵相结合的方法对滚动轴承早期轻度内、外圈故障的诊断问题进行研究。采用形态差值算子对故障信号进行滤波处理,差分熵提取滤波后信号在时域中的冲击,根据冲击突变点的时间间隔和特定故障信号在时域中的周期性冲击间隔相一致的思想来实现对滚动轴承早期轻度内、外圈故障的有效诊断,并利用仿真信号和滚动轴承实测故障信号对所提方法的有效性进行验证。(4)针对乘积分量的解调和单尺度形态学分析的结构元素长度选择的经验性和主观性问题,提出一种基于局部均值分解和改进自适应多尺度形态学分析相结合的方法。采用局部均值分解方法对信号进行分解,筛选若干个含有特征信息最丰富的乘积分量作为解调用数据源,基于扁平型、三角型和半圆型结构元素的改进自适应多尺度形态学分析方法对其进行解调分析,得到有效解调阈值区间(即在此区间,改进自适应多尺度形态学分析的解调效果要优于改进前自适应多尺度形态学分析)和最优解调阈值点。此外,将最优解调结果和改进前自适应多尺度形态学分析、Hilbert变换、Teager Kaiser能量算子和局部均值分解的解调结果进行对比,并利用液压泵实测故障信号对所提方法的有效性进行验证。(5)针对齿轮故障的诊断问题,提出一种基于局部均值分解和广义形态分形维数相结合的方法。采用局部均值分解方法对信号进行分解,筛选若干个含有最丰富特征信息的乘积分量作为数据源进行广义形态分形维数的提取,选取某些能够有效地表征齿轮运行状态非线性信息的维数作为特征向量,利用核模糊C均值聚类方法进行故障诊断。此外,还对信号长度、转速和负载扭矩对广义形态分形维数的影响进行分析。并利用仿真信号和齿轮实测故障信号对所提方法的有效性进行验证。(6)利用基于峭度、能量和均方差的选择方法对合理地选取含有特征信息丰富的乘积分量作为数据源的问题进行研究。采用局部均值分解方法对原信号进行分解,分别提取原信号和各个乘积分量的峭度、能量和均方差,并分别将这三个指标组成一个指标向量,计算原信号和各个乘积分量的指标向量之间的欧氏距离,选取若干个最小欧氏距离值所对应的乘积分量作为数据源,并利用液压泵实测故障信号验证所提方法的有效性。
[Abstract]:Rotating machinery parts, such as hydraulic pumps, rolling bearings and gears, play an important transmission role in industrial fields. They are widely used in metallurgical equipment, engineering machinery, precision machine tools, precision instruments, aerospace, automobile, ship and petrochemical industry. These parts are usually in high temperature, high pressure and high pressure. High speed and other bad working conditions have resulted in the rapid deterioration of their health state and high failure rate. In this paper, a method based on mathematical morphology and local mean decomposition is studied for the fault diagnosis of rotating machinery parts. The mathematical morphology and local mean mean decomposition are based on the characteristics of signal morphology. The analysis is carried out, so the two are highly adaptive and data driven. This paper mainly conducts the following research work: (1) aiming at the empiricism and arbitrariness of the length selection of the flat structure element, a combination of the morphological difference operator and the power spectral entropy is proposed to determine the most flat structural elements. The method of optimal filtering length is used to filter the signal with the morphological difference operator of the length of different flat structure elements. The power spectral entropy of the filtered signal is calculated. The length of the minimum value of the signal power spectrum entropy is the optimal filter length of the flat structure element, and the simulation signal and the hydraulic pump are used to measure the fault letter. The validity of the proposed method is verified. (2) a method based on the combination of morphological difference operator and morphological index is used to study the problem of the health state evaluation of the hydraulic pump with the slipper wear fault as the research object. The morphological difference operator is used to filter the fault signals of the slipper with different wear degrees. The morphological index of the fault signal under various degrees of wear is analyzed, and the change law and sensitivity of the slipper wear fault and its deterioration degree are analyzed by the morphological index. The diagnosis and deterioration degree of the slipper wear fault are evaluated, and the effectiveness of the proposed method is verified by using the measured fault signal of the hydraulic pump. (3) using a kind of shape based on the shape of the fault signal. The method of state difference operator and differential entropy is used to study the diagnosis of the early mild inner and outer fault of rolling bearing. The morphological difference operator is used to filter the fault signal, the differential entropy is used to extract the impact of the signal in the time domain, and the time interval of the shock point and the specific fault signal in the time domain. The idea of periodic impact interval is consistent to realize the effective diagnosis of early mild inner and outer fault of rolling bearing, and verify the validity of the proposed method using the simulation signal and the measured fault signal of the rolling bearing. (4) the empiricism of the structural element length selection for the demodulation of the product component and the single scale morphologic analysis. A method based on local mean decomposition and improved adaptive multi-scale morphological analysis is proposed. The local mean decomposition method is used to decompose the signal, and several product components with the most abundant feature information are selected as the demodulation data source, based on the flat, triangular and semicircular structure elements. An improved adaptive multiscale morphological analysis method is used to demodulate and analyze it, and the effective demodulation threshold interval is obtained (i.e., the demodulation effect of the improved adaptive multi-scale morphological analysis is better than the improved adaptive multiscale morphological analysis) and the optimal demodulation threshold point. Multiscale morphological analysis, Hilbert transformation, Teager Kaiser energy operator and local mean decomposition are compared, and the effectiveness of the proposed method is verified by using the measured fault signal of the hydraulic pump. (5) in view of the diagnosis of gear fault, a kind of local mean decomposition and generalized fractal dimension fractal dimension are proposed. The local mean decomposition method is used to decompose the signal and select the product components with the most abundant feature information as the data source to extract the generalized fractal dimension, and select some dimensions of the nonlinear information of the gear running state effectively as the feature vectors, and use the kernel fuzzy C mean clustering. In addition, the effect of the signal length, speed and load torque on the generalized fractal dimension is analyzed. The validity of the proposed method is verified by the simulation signal and the measured fault signal of the gear. (6) the selection method based on kurtosis, energy and mean square deviation is used to select the characteristic letter. The partial mean decomposition method is used to decompose the original signal and extract the kurtosis, energy and mean square of the original signal and the product components respectively, and the three indexes are composed of an index vector to calculate the original signal and the index vector of the product components respectively. Euclidean distance is used to select the product components corresponding to a number of minimum Euclidean distance values as the data source, and the effectiveness of the proposed method is verified by a hydraulic pump fault signal measured by a hydraulic pump.

【学位授予单位】：燕山大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH17

【相似文献】